You are on page 1of 8

Demand for Coal: The Problem

of Aggregation
Thomas H. Stevens, Martin J. Blake,
and Lawrence G. Williams
This paper demonstrates the differences that result from estimating coal demand
functions using highly aggregated regional data rather than less aggregated state data. At
first glance the coal demand functions based on regional data appear to explain well and
seem useful in policy making. However, coal demand functions based on state data show
differences in the demand structure in each state of the region. Policy decisions that may
seem appropriate based on aggregated regional data may have much different impacts on
individual states in the region.
Coal is a potential source of energy to
satisfy increased energy demand over the
next decade. Major adverse environmental
and economic impacts may, however, be as-
sociated with increased coal production.
Among these are: (a) land subsidence; (b) sur-
face disturbance; (c) water and air pollution
and (d) "boom town" phenomena.
In response to a general concern over the
relationship between increased coal de-
velopment, environmental quality, and the
demand for public services, several
econometric studies of the demand for coal
have recently been published. The results
indicate that the demand is price inelastic.
Therefore, it is concluded that more strin-
gent strip mine regulations or increased
severance taxes would have a negligible ef-
fect upon coal output and employment.
These conclusions are, however, based
upon econometric coal demand functions that
are estimated using highly aggregated data.
In this paper it is suggested that coal demand
functions based upon such data may not be
adequate for the formulation and evaluation
of public policy. As an example, a case study
of coal demand in the Four Corners Region
Thomas H. Stevens, Martin J. Blake and Lawrence G.
Williams are Assistant Professor, University of Massa-
chusetts, Assistant Professor, New Mexico State Univer-
sity and graduate student at New Mexico State Univer-
sity, respectively.
of New Mexico, Colorado, Utah, Arizona and
Nevada is developed. Conclusions for the
formulation of public policies are then pre-
sented and discussed.
Previous Research
Several coal demand studies have recently
been published [Goldstein and Smith; Libbin
and Boehlje; Lin, Spore and Nephew; and
Reddy]. As shown in Table 1, relatively con-
sistent estimates of price and cross price elas-
ticities have been obtained.
An analysis of these studies, revals, how-
ever, that coal demand contains a number of
features which are difficult to model. First,
the data are not extensive and are of rela-
tively low quality. Second, specification er-
rors may result due to the existence of price
regulation, electricity pooling arrangements,
and related factors. Of particular interest
here is that little attention has been devoted
to potential problems associated with aggre-
In order to provide a frame of reference, a
brief synopsis of each demand study cited
above is necessary. Reddy used a cross-
sectional data base consisting of a sample
which included only those steam-electric
generating plants equipped to utilize coal and
one or more of the other fossil fuels. The
demand for steam coal was postulated as a
Western Journal of Agricultural Economics
TABLE 1. Estimated Price and Cross Price Elasticities of Coal Demand
Study Region Price Elasticity Cross Price Elasticity
for Coal Natural Gas Oil
Reddy U.S.A. -. 371 to -. 974 .496 to 1.583 1.035 to 1.958
Lin, Spore
and Nephew Appalachia -1.65 .144 .216
and Smith U.S.A. -. 480 to -.316 -.937 to-.550 .283 to .260
function of coal price, oil price, gas price and
net generation of electricity by steam electric
generating plants. The generating plants
were aggregated according to size and fuel
combinations, and a separate demand equa-
tion was estimated for each grouping. The
results indicated price inelastic demand for
steam coal, with the elasticity values ranging
from -. 371 to -. 974.
Goldstein and Smith examined the price,
output and employment effects of increased
land reclamation requirements. Coal used for
electricity generation was aggregated with
that used for steel production. The results
indicated that reclamation standards would
have only slight effects upon output and em-
Lin, Spore and Nephew (LSN) also focused
upon the probable impacts of increased rec-
lamation requirements on mining costs, coal
prices, production and employment. The
demands for steam, coking, industrial and
export coal were estimated separately. The
resulting demand for steam coal was both
price and cross-price inelastic. The price in-
elasticity was attributed to difficulty of short
run interfuel substitution. However, the
prices of fuel inputs were weighted by their
consumption in order to account for the ef-
fect of oil and gas price regulation. Moreover,
the price of coal was weighted by steam elec-
tric generation in each state of the region.
Consequently, the respective parameter es-
timates cannot be readily interpreted as elas-
ticities. That is, the ratio representing the
price variables may change, thus indicating a
change in the demand for coal, but the result-
ing change in demand cannot be attributed
solely to a change in the price of coal or the
substitute fuel.
Of greater importance is that the LSN de-
mand model contained variables for prices of
coal, oil, and natural gas. Yet, Gordon [1975,
1976] found that "No more than two fossil
fuels have been major potential suppliers in
any one region." In the Appalachian region,
the two major competing fuels differ from
state to state. Aggregating state data with re-
spect to all three fuels, when in essence one
fuel may not be available in any one specific
state, may result in a misspecified model.
In summary, each of these models em-
ployed a different type and degree of aggre-
gation. Moreover, each used highly aggre-
gated regional or national data. It is
hypothesized here that such models may not
be adequate for the formulation or evaluation
of policy from a state or local perspective.
This inadequacy is partially because the
structures of both the electrical and coal in-
dustries differ among states and regions.
The Demand Models
For the purpose of testing this hypothesis,
the demand for coal in the Four Corners Re-
gion was estimated using two different
econometric models. The first was a single
coal demand function representing the entire
Four Corners Region while the second con-
tained separate coal demand functions for
each state within the region.1
The Regional Demand Model
The regional econometric coal demand
model contained two fuel price variables: coal
and natural gas.
The total quantity of elec-
New Mexico, Colorado, Utah and Arizona and Nevada
July 1979
Coal Demand and Aggregation
tricity generated within the region was in-
cluded as a third independent variable. The
model was specified as:
lnQs = Bo + B
lnPc + B
lnPg + B3lnE + ut.
Where Qs = quantity of coal consumed for
electric power generation, in thousand tons;
Pc = average price of coal as consumed by
electric utilities, in cents per million BTU
and deflated using the GNP implicit price
deflator; Pg = average price of natural gas as
consumed by the electric utilities, in cents
per million BTU and deflated using the GNP
implicit price deflator; E = total kilowatt
hours of electricity generated in the Four
Corners Region; and ut = an error term.
This specification assumes strong separa-
bility in the production of energy (BTU's)
such that: BTU = F
(C) + F2(G) where C and
G represent coal and natural gas, respec-
tively. It also assumes that there are no sub-
stitutes for energy (BTU's) in the production
of electricity. As shown by Goldstein and
Smith, this implies that the production of
electricity, E, requires a certain quantity of
BTU = f(E). The quantity of BTU's con-
sumed is, therefore, a function of output
alone and not of labor or capital prices. We
also assume that the quantity of electricity
produced, E, is independent of fuel prices
[Goldstein and Smith, Reddy]. To the extent
that these assumptions are violated, a simul-
taneous equation system specification would
be required. However, since our objective is
to explore the problem of geographical data
aggregation, a model specification consistent
with that used by Goldstein and Smith, Lin,
Spore and Nephew, and Reddy was utilized
in this study.
Aggregated time series data were not
readily available for the region. Therefore,
the individual state data for the 1963-74
period were aggregated to obtain data for the
entire region, giving a total of 12 observa-
Historically, coal-gas competition has prevailed within
the region. Although fuel oil was extensively used in
Utah prior to 1973, Utah produces less than six percent
of the total energy generated within the region.
tions. The ordinary least squares (OLS) re-
sults were as follows:
lnQ,s = -7.1610
- .4000 lnPc
(- 1.011)
+ .3053lnPg +
1.5407 InE
The values in parentheses are calculated
t-values and the values in brackets are
standard errors of the estimates. The
electricity-generating parameter estimate
was significant at the .01 level while the
natural gas price and the coal price variables
were significant at approximately the .30
level. The R
value equaled .984 and the
Durbin-Watson (DW) statistic was 1.68. The
Durbin-Watson test was, therefore, incon-
clusive regarding the existence of autocorre-
lation of the disturbance terms at the 95 per-
cent level.
The results indicate both price and cross-
price inelasticity. These findings are gener-
ally consistent with the studies reviewed
above (Table 1). From a macro (regional)
perspective it, therefore, appears that
policy-induced mining-cost increases may
have a relatively small effect upon coal out-
put or employment. However, this inference
may not be valid from a micro (state or local)
perspective for several reasons.
First, regional average price data obtained
from the Bureau of Mines were used. De-
pending on the size of coal shipments, trans-
portation costs and hence prices could vary
substantially within the region and from plant
to plant. Consequently, the impact of policy-
induced increases in the average price of coal
may vary substantially within the region.
Second, the data on average coal price in-
clude both contract and spot prices. The av-
erage price, therefore, may not represent the
market clearing price for any individual state
or firm. Finally, the structure of the indus-
try varies within the region. For example,
although coal-gas competition has historically
Stevens, Blakce, and Williams
Western Journal of Agricultural Economics
prevailed within the region, fuel oil was ex-
tensively used in Utah prior to 1973.
Individual State Demand Functions
In order to investigate the implications of
using aggregated regional data, separate de-
mand functions were estimated for each of
the states within the region (Arizona and
Nevada combined, Colorado, New Mexico,
and Utah). Disaggregation to a higher degree
would be desirable, but was not possible due
to the lack of available data. The OLS esti-
mates for each state are given in Table 2.
As shown in Table 2, the R
values are high
for each relationship, but two parameter es-
timates exhibited the wrong sign and several
were statistically insignificant. Specifically,
the price of coal was found to be insignificant
and to have the wrong sign in both the
Arizona-Nevada and Utah functions.
Durbin-Watson test indicated no significant
autocorrelation for the Arizona and Nevada
and New Mexico models at the 95 percent
level, with inconclusive results for Utah and
Moreover, the effect of aggregation may
depend not only upon the aggregation proce-
dures adopted, but also upon the method of
statistical estimation. For coal demand, the
OLS technique may generate unreliable
and/or inefficient parameter estimates for
each state because of interactions in industry
structure which are not reflected by the in-
dependent variables in the model. For
example, an electricity pooling arrangement
comprising two or more states would be sub-
ject to control by each state's regulatory
Pooling cross section and time series data
using seemingly unrelated regressions is one
way to account for the effects of such unquan-
tifiable factors in the parameter estimates.
Seemingly unrelated regressions estimation
is appropriate when the variables that are ne-
glected in a set of equations are partly the
same or to some degree correlated. This
would result in a correlation of the distur-
bance terms among the equations. Although
High R
values associated with insignificant parameter
estimates may result if significant multicollinearity
exists. Multicollinearity did not appear to be substantial
in the Arizona-Nevada function. Multicollinearity be-
tween the price of coal and the price of gas was, how-
ever, present in the Colorado function. This may ex-
plain the high standard error associated with the coal
price variable in Colorado.
TABLE 2: Estimated Coal Demand Equations, OLS Regression Technique,a
State Estimated Equation R
Arizona & Nevadad In Qsc = -21.56 + .23 In P + 1.08 In Po + 2.44 In E .98 2.05
(- 7.49) ( .46)
(1.92) (18.21)
[ 2.87] [ .49] [ .56] [ .13]
Colorado In Qsc= - 1.72 - .82 In Pc + 1.01 In Pg + .99 In E .93 1.24
(- .51)
(- .62)
(1.38) ( 6.74)
[ 3.41] [ 1.32] [ .73] [ .15]
New Mexico In Qsc = 0.31 - 1.03 InPc+ 0.15 In Pg + 1.11 In E .99 1.91
( .12)
(-1.73) ( .62)
( 6.61)
[ 2.66] [ .59] [ .25] [ .17]
Utah In Qsc = -12.49 + .45 In Pc + .45 In Pg + 1.95 In E .85 1.50
(- 2.09) ( .71)
(2.18) ( 2.77)
[ 5.98] [ .62] [ .21] [ .70]
aFigures in parentheses are calculated t-values.
bFigures in brackets are standard error values.
Clnsignificant at the a = .20 level.
dPrice of oil used as substitute fuel (PO).
July 1979
Coal Demand and Aggregation
the equations seem unrelated, they are in
fact disturbance-related [Theil]. In the case
of coal demand in the Four Corners Region,
the disturbance terms may be correlated be-
cause of similar institutional relationships,
regulations and other common factors that af-
fect all states in the region. Thus, seemingly
unrelated regressions estimation should give
more efficient estimates of the parameters in
this case.
The results obtained from the seemingly
unrelated regressions technique are present-
ed in Table 3. This technique produced more
efficient parameter estimates than did OLS,
as evidenced by the lower standard errors on
the estimated coefficients and by only one
sign that contrasted with theory. Given the
results in Table 3, the demand estimates
based on regionally aggregated data may not
be appropriate for policy evaluation. For
example, the coal price elasticity of demand
is estimated to be -1.33 in New Mexico,
-1.52 in Colorado and -. 07 in Utah (Table
3). These do not closely correspond to the
regional price elasticity estimate of -. 40
based on aggregated data.
The results tend to indicate that policies
formulated from a regional perspective may
cause highly variable responses from specific
areas within the region. That is, a policy may
initially look "good" for the region but turn
out "bad" for almost every state within the
Impacts of Using
Regionally Aggregated Data
It is important to realize that these two
quite different sets of results are based on the
same data. Individual state data were
summed to get the regional quantity of coal
consumed and total kilowatt hours of electric-
ity generated. A weighted average of indi-
vidual state prices was taken to obtain the
regional average price of coal and the average
price of natural gas. The reason that the re-
gional model seems to have greater explana-
tory ability than the separate state models is
that the variability of the data used in es-
timating the regional model has been re-
duced in the process of aggregation. Some of
the individual differences reflected in the
state data tend to cancel when aggregated
into regional figures.
Equations estimated using aggregated data
may seem to perform better simply because
there is less variation to explain. However,
using aggregated data can be harmful if the
estimated equations are used as a basis for
policy decisions. The individual differences
often are important in policy formulation.
For example, an increase in reclamation re-
quirements may lead to only a slight reduc-
tion in regional output. However, output
may fall significantly in New Mexico and
Colorado and very little in Arizona and
Nevada and Utah.
TABLE 3: Estimated Coal Demand Equations Using the "Seemingly Unrelated Regression"
State Estimated Equation
Arizona & Nevada In Qsc = -20.35 - 0.01 In Pc + 0.72 In Pg + 2.52 In E
[ 2.68] [ .45] [ .52] [ .12]
Colorado In Qsc = - 0.39 - 1.52 In Pc + 1.33 In Pg + .98 In E
[ 2.94] [1.11] [ .63] [ .13]
New Mexico In Qc = 1.59 - 1.33 In Pc + .22 In Pg + 1.04 In E
[ 2.53] [ .56] [ .24] [ .16]
Utah In Qsc = -10.77 - .07 In Pc + .57 In Pg + 1.89 In E
[ 4.97] [ .48] [ .17] [ .60]
aFigures in brackets are standard error values
and "t" tests are not comparable with OLS estimates.
Stevens, Blake, and Williants
Western Journal of Agricultural Economics
That is, the structure of the electrical in-
dustry in each state comprising a region may
be quite different. In fact, in this case,
Chow's test was used to determine whether
the state estimates can be regarded as be-
longing to the same regression model. The
results indicated that the relationships are
not the same at the 95 percent confidence
Thus, the structure of the electrical indus-
try in these states is basically different. A pol-
icy that may work well on one structure may
work poorly on another. Alternatively, a pol-
icy that seems appropriate for an average
structure may perform poorly in terms of the
structure's individual components.
Coal demand functions based on nationally
or regionally aggregated data, such as those
presented in the previous literature, may not
be adequate to describe specific situations
within a region. Even though a policy may
not appear to significantly affect coal output
at the regional level, significant shifts within
the region may result. Evidence has been
provided which indicates that econometric
coal demand functions based on regionally
aggregated data may be of little value in de-
termining the effects of coal-related policies
upon the states in the Four Corners Region.
Yet as Grunfeld and Griliches point out,
"in practice we do not know enough about
microbehavior to be able to specify micro-
equations perfectly. Hence, empirically es-
timated microrelations... (in this case indi-
vidual states) ... should not be assumed to be
perfectly specified... Aggregation of eco-
nomic variables can and, in fact, frequently
does, reduce these specification errors.
Hence aggregation... may provide an
aggregation gain." Consequently, further re-
search is required on appropriate micromod-
el specification and on the type of aggrega-
tion employed. As demonstrated here, such
research may substantially improve the qual-
ity of information utilized for policy formula-
tion and evaluation.
The specific implications of the different
sets of demand functions presented may be
summarized in several ways. First, environ-
mental quality legislation which results in in-
creased coal prices will tend to have the
greatest impact upon coal output and em-
ployment in Colorado and New Mexico.
Consequently, geographical variations in en-
vironmental policy may be warranted. Sec-
ond, the results provide a partial justification
for existing variations in coal severance tax
rates between the states within the region.
Third, the short-run effect of natural gas
price deregulation upon the coal industry is
likely to be relatively insignificant as indi-
cated by the cross-price elasticity values re-
ported in Tables 2 and 3.
The principal policy implication is, how-
ever, that econometric estimates based upon
highly aggregated data may fail to account for
economic and physical characteristics unique
to each geographical area within the aggre-
gate. Evidence has been provided which in-
dicates that the states within the Four Cor-
ners Region must be treated as separate
entities in the formulation of coal develop-
ment policies. If this is not done, significant
redistribution may result within the region.
Geographical variation in coal development
policies is, therefore, advocated as a means of
avoiding or enhancing redistributional con-
Edison Electric Institute, Statistical Year Book of the
Electric Utility Industry, New York, (1963-1974 is-
Federal Energy Administration. Final Report, Short-
term Coal Forecast, 1975-1980, Washington: Office of
Coal, August 1975.
, Project Independence Report, U.S. Government
Printing Office, Washington, D.C., November 1974.
Goldstein, Morris, and Robert S. Smith, "Land Recla-
mation Requirements and Their Estimated Effects on
the Coal Industry," Journal of Environmental Eco-
nomics and Management, December 1976.
July 1979
Coal Demand and Aggregation
Gordon, Richard L., Marketing Prospects for Western
Coal, Pennsylvania State University, prepared for the
National Science Foundation, December 1976.
-- , U. S. Coal and the Electric Power Industry, Re-
sources of the Future, Washington, D.C., Johns
Hopkins University Press, 1975.
Grunfeld, Y. and Z. Griliches, "Is Aggregation Necessar-
ily Bad?," Review of Economics and Statistics,
42(1960): 1-13.
Libbin, James D. and Michael D. Boehlje, "Interregion-
al Structure of the U.S. Coal Economy," American
Journal of Agricultural Economics, 59(1977): 456-466.
Lin, William, Robert L. Spore, and Edmund A.
Nephew, "Land Reclamation and Strip-Mined Coal
Production in Appalachia," Journal of Environmental
Economics and Management, October 1976, pp. 236-
Reddy, Nallapu N., "The Demand for Coal: A Cross-
Sectional Analysis of Multi-Fuel Steam Electric
Plants," Industrial Organization Review, University
Publications Division of Physical Biological Sciences
Ltd., Volume 3, Number 1, 1975.
Theil, Henri, Principles of Econometrics, John Wiley
and Sons Inc., New York, 1971.
U. S. Department of Interior, Bureau of Mines, Minerals
Yearbook, Volume on Fuels, Washington, D.C., U.S.
Government Printing Office, (1953-1974 issues).
Walsh, R. G., Some Benefits and Costs of Strip Mining
Western Coal Resources, Las Cruces, New Mexico,
Department of Agricultural Business and Agricultural
Economics, New Mexico State University, April 1974.
Stevens, Blake, and Williams
Western Journal of Agricultural Economics
July 1979